Autonomic Distribution and Adaptation

This chapter describes an approach for increasing the scalability of applications by exploiting inherent concurrency in order to parallelize and distribute the code. We thereby focus specifically on concurrency in the sense of reduced dependencies between logical parts of an application. Concurrency forms a crucial part in any par- allelization approach, as the degree of dependencies across potential threads defines the delay due to messaging and synchronization overhead. For example, loop unroll- ments show best performance improvement if they are highly concurrent and thus vectorizable.